AI for software companies: where the economic leverage really is

For software companies the interesting question is not whether AI helps with programming. It does. The interesting question is where in the value chain the effect actually lands – and why many firms see more code but barely better margins after a year of using AI.

1. The starting point

A software company sells project outcomes, not lines. Commercial success hangs on three quantities: how much effort a project ties up, how predictable that effort is, and how much of it stays reusable in follow-up projects. AI affects all three – but to very different degrees.

2. What a project really consists of

A rough breakdown of a custom project that many firms will recognise:

The uncomfortable observation: the share with the lowest perceived client value ties up the largest predictable effort in many projects – and it repeats almost identically in every project.

3. Three levers, very different effects

1. Type faster. Affects implementation, so part of the effort. Immediately noticeable but limited – and, without process changes, partly eaten up by extra review effort.
2. Have to type less. Affects the technical base: whatever exists as a platform capability needs neither generating nor reviewing nor maintaining. The effect is larger and lasting.
3. Be able to take on more demanding projects. Affects revenue rather than cost – for instance when regulated or evidence-heavy projects become feasible that previously failed on the cost of producing evidence.

4. Reuse, accelerate, control

A simple three-way split for project planning follows from this:

• Reuse what is already solved: identities, roles and permissions, tenants, data access, history, reporting foundations. Which parts sensibly belong there is covered in the deep dive platform instead of boilerplate.
• Accelerate with AI what stays bespoke: domain logic, interfaces, integrations. Speed pays off directly here, because this is the part that carries the client value.
• Control what ships: reviews, quality gates, traceability, reproducible delivery. That is the part that turns fast work into a product you can stand behind.

5. What this means for costing

When the technical base no longer arises project by project, the cost structure changes: part of the variable project effort becomes a fixed platform item. That is commercially attractive, but only under two conditions:

• Utilisation. A shared base pays off from a certain number of parallel or consecutive projects – with a single project a year almost nothing pays off.
• Discipline. As soon as projects fork the base “just a little”, per-project maintenance returns and the effect reverses.

Concretely for proposals: the base share moves out of the effort estimate and into a platform or licence line item. The estimate becomes smaller and more reliable – and the second effect is the more valuable one over time.

6. Proposals and client communication

Clients now ask about AI use directly. Two answers work badly: “we do not use AI” (often untrue) and “it makes us 50% faster” (unprovable). The third variant holds up: describe how it is used and how it is checked.

That means a short statement on the operating model, on how client code is handled and on the checks before delivery. The basis for it is a written policy – see AI coding governance. Answering this cleanly in a proposal is an advantage in tenders over competitors who dodge the question.

7. Risks of the wrong priority

• Price pressure without differentiation. When everybody implements faster, competition shifts to price – unless the offer contains something that cannot be generated: domain knowledge, the ability to evidence, responsibility for operations.
• Maintenance load. More code per project means more code you maintain for years. Speed without standardisation produces exactly that.
• Quality risk at scale. When every team has its own rules, the claim “our projects meet standard X” no longer holds.

8. A realistic way in

Checklist: pulling the right lever

Conclusion

For software companies, AI shifts not primarily the speed but the question of which work still has to be project-specific at all. The leverage is in the combination: a reusable base, AI for the bespoke part, and a controlled path to delivery.

Tackle only the middle part and you get faster – and stay in the same cost structure. Tackle all three and the arithmetic changes.